Optimize RNG, visited tracking, and arrow representation

[Nanosplitter]

Summary

This PR refactors core performance-critical components of the simulation engine, replacing slow standard library functions with optimized implementations and improving data structure efficiency.

Key Changes

• Replace rand() with xorshift32 PRNG: Implemented a fast inline xorshift32 pseudo-random number generator with unbiased range selection via multiply-shift, eliminating the bias and overhead of rand() % b

• Optimize visited state tracking: Replaced boolean arrays with a stamp-based approach using uint16_t counters. Instead of clearing the entire visited table each run, a monotonically increasing stamp marks visited states. This avoids O(HEIGHT×WIDTH×NUMDIR) memset operations on every simulation run, with automatic wraparound handling

• Introduce Arrow struct: Replaced vector<int>{y, x, dir} with a trivially copyable POD struct. This provides:

  • Type-safe field access (.y, .x, .dir instead of .at(0), .at(1), .at(2))
  • Contiguous memory layout for vectors
  • Allocation-free copying
  • Bounds-check-free indexing

• Simplify arrow generation: Refactored genRandArrows() to use partial Fisher-Yates shuffling (O(numRandArrows) instead of O(validCells.size())), eliminating unnecessary full-array shuffles

• Remove unused code: Deleted getArrString() from Simulation class, unused copyUsedCells() method, and numerous commented-out debug statements

• Improve code clarity: Changed generateNewArrows() to modify in-place (void return) rather than returning a copy, simplified array initialization with memset, and replaced verbose loops with cleaner implementations

Implementation Details

• The stamp-based visited tracking wraps around at 65535 runs, triggering a single reset to avoid overflow
• Arrow struct enables direct array indexing without bounds checks while maintaining type safety
• Partial Fisher-Yates reduces random number generation calls and array operations during initialization

https://claude.ai/code/session_01LRo9aeU3LgH3nrfwxtZqMA